Not Just Reason, Not Just Scan: Reinforcement Learning for Proactive Scientific Error Verification over Academic Paper

Rongjin Li; Yuanxin Liu; Hao Zhou; Fandong Meng; Jie Zhou; Xu Sun

cs.CLAug 27, 2026

Not Just Reason, Not Just Scan: Reinforcement Learning for Proactive Scientific Error Verification over Academic Paper

Authors: Rongjin Li, Yuanxin Liu, Hao Zhou, Fandong Meng, Jie Zhou, Xu Sun

Organizations: Beijing University of Posts and Telecommunications · State Key Laboratory for Multimedia Information Processing, School of Computer Science, Peking University · WeChat AI, Tencent Inc., China

Abstract

Multimodal large language models (MLLMs) are increasingly capable scientific assistants, yet they remain far from fully autonomous research. This transition requires models to actively inspect academic papers, build global evidence views, and make traceable judgments without prespecified issues or evidence. However, existing work provides limited task paradigms or training studies for such issue- and evidence-absent verification. We study this challenge through scientific error detection, where models must determine whether errors exist and justify them with evidence-based reasoning. To fill this gap, we present VERA-RL, a reinforcement-learning formulation for scientific error detection over academic papers. Following a Reason--Verify--Scan progression, we construct VERA-13K, a 12,900-sample dataset organized into 4,300 matched chains, covering 6 scientific-error categories across the research workflow and broad natural-science domains. We further introduce fine-grained rewards for reasoning completeness, evidence alignment, and error precision. Training Qwen3-VL-8B with VERA-RL substantially improves verifiable reasoning, approaching flagship MLLMs such as Gemini 3 Pro and Qwen3-VL-235B-A22B on Scan.

Figures & tables

Figure 1: Overview of Reason - Verify - Scan paradigm.

Figure 2: Sampled VERA-13K examples with 6 error types. We consolidate and extend the taxonomy proposed in ScholScan, resulting in a more coherent and well-structured categorization scheme.

Statistics	Train Dataset		Test Dataset
Statistics	SFT	RL	Avg.	QI	DI	IC	PD	RQD	SG
# Examples	10,500	1,500	-	150	150	150	150	150	150
Avg. Length	31,134	30,649	31,837	29,956	37,066	31,985	31,100	32,792	28,124
Avg. Evidence	3.54	3.52	3.58	3.28	3.28	3.80	3.60	3.92	3.62
Avg. Reasoning Step	3.81	3.79	3.79	3.77	3.85	3.72	3.84	3.81	3.74

Table 1: Detailed statistics and distributions of our train and test datasets. Length is calculated by the Qwen3 tokenizer. Scan , Verify , and Reason are strictly balanced in every split and are constructed as a chained curriculum from the same underlying instances. QI : Quantitative Inconsistency; DI : Design & Identifiability; IC : Inference & Conclusions; PD : Pipeline Distortion; RQD : Research Question & Definitions; SG : Sampling & Generalizability.

Question Prompt: Read the provided paper and answer the question. The question will ask you to check and reason about scientific errors in specific parts of the paper. In your answer, show your reasoning process and explain in detail the exact nature of the errors. <Question> <Interleaved context of the paper>

Gold Answer: Evidence: (Only for Scan questions) - <Evidence 1> - ... - <Evidence n > Reasoning: - <Step 1> - ... - <Step n > Answer: ...

Table 2: QA format used in VERA-13K .

Models	Avg.	QI	DI	IC	PD	RQD	SG
Scan $R_{\mathrm{completeness}}$
Gemini 3 Pro	22.8	24.4	13.2	30.0	17.7	25.8	26.0
Qwen3-VL-235B-A22B (Thinking)	16.7	22.2	12.5	21.7	10.2	13.3	20.3
Qwen3-VL-235B-A22B (Instruct)	3.4	7.0	3.3	7.8	1.5	0.0	0.5
Qwen3-VL-8B (Instruct)	1.5	0.0	0.0	5.3	1.5	0.0	2.0
Qwen3-VL-8B (SFT, ours)	5.3	5.2	1.3	5.8	6.3	3.0	10.3

Table 3: Scan -task evaluation results (scaled by 100) for baselines and models trained in the main experiments.

Models	$R_{\text{completeness}}$	$R_{\text{alignment}}$	$S_{\text{reason}}$	$S_{\text{location}}$	$P_{\text{unrelated\_err}}$	$S(m)$
Gemini 3 Pro	11.3	7.2	11.0	6.0	10.2	5.5
Qwen3-VL-235B-A22B-Thinking	4.7	2.8	4.6	2.4	5.3	2.6
Qwen3-VL-235B-A22B-Instruct	0.8	0.4	0.8	0.3	0.4	0.1
Qwen3-VL-8B-Instruct	0.0	0.0	0.0	0.0	0.0	0.0
Qwen3-VL-8B (SFT)	0.4	0.1	0.4	0.0	3.8	0.0
Qwen3-VL-8B (SFT+RL)	0.5	0.3	0.5	0.2	0.5	0.2

Table 4: Evaluation results (scaled by 100) for baselines and models trained in the main experiments.

S_{\text{reason}}

S_{\text{location}}

P_{\text{unrelated\_err}}

, and

S(m)

are evaluation metrics introduced in ScholScan, with detailed definitions provided in Appendix C .

R_{\text{precision}}

is not applicable in this evaluation setting and is therefore omitted.

Figure 3: Left : Training dynamics of main experiments. Right : Metrics change after VERA-RL on Scan ( test set).

Figure 4: Performance comparison between multi-dimensional and single reward configurations.

Figure 5: Training dynamics of the ablation study using only

R_{\mathrm{completeness}}

Config.	Step	Reason			Verify			Scan
		$R_{\text{final}}$	$R_{\text{com}}$	$R_{\text{precision}}$	$R_{\text{final}}$	$R_{\text{com}}$	$R_{\text{precision}}$	$R_{\text{final}}$	$R_{\text{com}}$	$R_{\text{align}}$	$R_{\text{precision}}$
Main	30	38.5	17.6	69.9	37.7	17.2	68.5	19.5	8.2	6.2	68.6
Main	15	36.1	14.6	68.4	35.0	14.9	65.2	15.0	2.8	2.3	65.0
Pure Scan	10	37.0	16.7	67.4	34.0	12.8	65.8	16.8	5.3	3.7	66.0
$\Delta$ to Main (30)	-	-1.5	-0.9	-2.5	-3.7	-4.4	-2.7	-2.7	-2.9	-2.5	-2.6
Pure Scan	20	37.2	15.9	69.1	34.1	12.6	66.3	16.9	5.3	4.2	65.4

Table 5: Performance comparison across different configurations for Reason , Verify , and Scan tasks. The metric

R_{\text{com}}

refers to

R_{\text{completeness}}

and

R_{\text{align}}

refers to

R_{\text{alignment}}

. The Pure Scan configuration isolates the Scan task for training.

Figure 6: Training dynamics of the ablation experiment using only Scan tasks.

Appendix figures & tables20 assets

Supplementary material from the paper’s appendix.

Appendix

Benchmark	Modality	Paradigm	Eval.	Domains	Count (K)
CharXiv	I	Reason	Close	8	11.6
ArXivQA	I	Reason	Close	10	100
MMCR	T+MD	Reason	Close	CS	310
AAAR-1.0	T+MD	Reason	Mixed	CS	13.5
PRISMM-Bench	I	Reason	Close	CS	0.4
ScholScan	T+MD	Scan	Open	13	1.8

Appendix

Table 6: Comparison with representative academic document understanding benchmarks. T : text; I : image; MD : multimodal document; CS : Computer Science.

Figure 7: Distribution of paper sources.

Domain	Count
Life Sciences	227
Medicine	107
Materials	67
Ecology	34
Chemistry	32
Environment	30

Appendix

Table 7: Broad-domain distribution beyond Computer Science.

Subdomain	Count
Foundation Models & LLM	424
Computer Vision	412
Multimodal Learning	287
Generative Modeling	243
Reinforcement Learning	194
Evaluation	190

Appendix

Table 8: Major AI/ML subdomain distribution. Counts are paper-level statistics, and only the most represented subfields are listed.

Category	Count
QI	3450
DI	540
IC	2190
PD	1470
RQD	3450
SG	900

Appendix

Table 9: Train-subset distribution across 6 scientific-error categories.

Split	Total	Review-derived	Controlled-edit	Review (%)
SFT Train	10,500	2,856	7,644	27.2
RL Train	1,500	492	1,008	32.8
Test	900	276	624	30.7
Total	12,900	3,624	9,276	28.1

Appendix

Table 10: Source composition of VERA-13K. Review-derived samples are constructed from objective errors extracted from peer reviews, while controlled-edit samples are generated by injecting errors into accepted papers.

Models	Avg.	QI	DI	IC	PD	RQD	SG
Scan $R_{\mathrm{completeness}}$
GPT-5.4	39.5	54.3	42.6	33.0	36.8	37.7	32.9
Qwen3-VL-Plus	20.6	34.2	7.5	28.5	16.0	19.2	18.3
Qwen3-VL-32B	12.7	9.0	6.7	19.5	10.0	13.8	17.3
Seed-1.6-Thinking	13.4	19.1	11.2	15.7	8.8	14.0	11.5
Scan $R_{\mathrm{alignment}}$

Appendix

Table 11: Supplementary sub-metric scores on the Scan task (scaled by 100).

Models	Avg.	QI	DI	IC	PD	RQD	SG
Reason $R_{\mathrm{completeness}}$
GPT-5.4	71.1	82.1	62.8	58.4	78.6	72.0	71.9
Gemini 3 Pro	59.2	64.7	55.3	63.0	57.0	65.2	49.9
Seed-1.6-Thinking	52.4	58.4	49.8	52.0	58.4	59.6	36.3
Qwen3-VL-Plus	60.7	68.8	52.3	68.0	60.8	64.8	49.6
Qwen3-VL-235B-A22B (Thinking)	54.7	67.1	42.7	64.0	58.1	55.2	40.9

Appendix

Table 12: Sub-metric scores on the Reason task (scaled by 100).

Models	Avg.	QI	DI	IC	PD	RQD	SG
Verify $R_{\mathrm{completeness}}$
GPT-5.4	62.3	74.7	51.2	67.6	65.5	71.0	49.7
Gemini 3 Pro	52.3	68.8	40.5	46.3	62.4	55.0	40.5
Seed-1.6-Thinking	32.8	37.3	22.0	27.2	39.9	42.2	28.4
Qwen3-VL-Plus	46.4	64.2	36.7	43.0	52.1	46.9	35.7
Qwen3-VL-235B-A22B (Thinking)	41.0	57.6	15.8	42.0	50.3	47.5	32.5

Appendix

Table 13: Sub-metric scores on the Verify task (scaled by 100).

Benchmark	Instruct	SFT+RL
MMLongBench-Doc	22.9	23.3
MMMU	44.3	46.1
PRISMM-Bench	51.6	52.0

Appendix

Table 14: Additional external evaluation results.

Model	Split	Reason	Verify	Scan
Instruct	Full	29.5	5.7	2.0
Instruct	Disjoint	30.2	5.2	1.6
SFT	Full	30.3	28.6	14.4
SFT	Disjoint	31.3	28.6	14.3
SFT+RL	Full	38.5	37.7	19.5
SFT+RL	Disjoint	39.3	39.1	19.3

Appendix

Table 15: Performance on the full and paper-disjoint test sets. All scores are

R_{\mathrm{final}}

and scaled by 100.

Metric	0 step	200 steps	1 epoch
Scan $R_{\mathrm{completeness}}$	1.5	1.6	5.3
Scan $R_{\mathrm{alignment}}$	1.0	2.4	2.5
Scan $R_{\mathrm{precision}}$	5.0	61.4	56.3
Scan $R_{\mathrm{final}}$	2.0	13.8	14.4
Verify $R_{\mathrm{final}}$	5.7	29.2	28.6
Reason $R_{\mathrm{final}}$	29.5	31.1	30.3

Appendix

Table 16: Performance of the intermediate SFT checkpoint at step 200, compared with the original base model and the 1-epoch SFT model. Scores are scaled by 100.

Step	Reason		Verify		Scan
Step	$\sigma(R_{\text{com}})$	$\sigma(R_{\text{prec}})$	$\sigma(R_{\text{com}})$	$\sigma(R_{\text{prec}})$	$\sigma(R_{\text{com}})$	$\sigma(R_{\text{align}})$	$\sigma(R_{\text{prec}})$
1	0.36	0.16	0.42	0.21	0.34	0.28	0.19
5	0.24	0.14	0.34	0.20	0.28	0.23	0.17
10	0.29	0.21	0.26	0.19	0.20	0.15	0.13
15	0.32	0.17	0.32	0.18	0.28	0.24	0.14
20	0.35	0.17	0.37	0.23	0.29	0.29	0.14

Appendix

Table 17: Standard deviation of reward components over rollout steps.

Model	Default			Completeness-heavy			Equal R/V
Model	Reason	Verify	Scan	Reason	Verify	Scan	Reason	Verify
Gemini 3 Pro	60.0	54.5	24.3	59.8	54.0	24.8	60.2	55.1
Qwen3-VL-235B-A22B Thinking	51.8	46.8	17.4	52.5	45.4	17.8	51.0	48.3
Qwen3-VL-235B-A22B Instruct	45.9	14.2	3.4	46.1	13.9	3.5	45.7	14.6
Qwen3-VL-8B Instruct	29.5	5.7	2.0	28.0	5.3	2.1	31.1	6.2
Qwen3-VL-8B SFT	30.3	28.6	14.4	26.2	24.7	14.7	34.4	32.7

Appendix

Table 18: Post-hoc reward reweighting results. Scores are scaled by 100. Default weights are

(0.6,0,0.4)

for Reason / Verify and

(0.4,0.4,0.2)

for Scan . Completeness-heavy weights are

(0.7,0,0.3)

for Reason / Verify and

(0.5,0.3,0.2)

for Scan . Equal R/V uses

(0.5,0,0.5)

for Reason / Verify .

Figure 8: Case study on inference and conclusions.

Figure 9: Case study on the first part of pipeline distortion.

Figure 10: Case study on the second part of pipeline distortion.

Figure 11: Case study on the first part of quantitative inconsistency.

Figure 12: Case study on the second part of quantitative inconsistency.

Setup	Reason ( $R_{\mathrm{final}}$ )	Verify ( $R_{\mathrm{final}}$ )	Scan ( $R_{\mathrm{final}}$ )
Qwen3-27B	78.3	88.4	88.7
Gemini 2.5 Flash	84.4	82.3	82.2

Appendix

Table 19: Correlation (%) between reward scores extracted by alternative evaluators and those from Seed-1.6-Thinking on the step-30 RL rollouts.

Language Model Error Detection

Language model error detection identifies unreliable answers, reasoning steps, and actions so systems can flag, verify, or repair them.

See all 90 papers

Mar 27, 2026

Not Search, But Scan: Benchmarking MLLMs on Scan-Oriented Academic Paper Reasoning

With the rapid progress of multimodal large language models (MLLMs), AI already performs well at literature retrieval and certain reasoning tasks, serving as a capable assistant to human researchers, yet it remains far from autonomous research. The fundamental reason is that current work on academic paper reasoning is largely confined to a search-oriented paradigm centered on pre-specified targets, with reasoning grounded in relevance retrieval, which struggles to support researcher-style full-document understanding, reasoning, and verification. To bridge this gap, we propose ScholScan, a new benchmark for academic paper reasoning. ScholScan introduces a scan-oriented task setting that asks models to read and cross-check entire papers like human researchers, scanning the document to identify consistency issues. The benchmark comprises 1,800 carefully annotated questions drawn from nine error categories across 13 natural-science domains and 715 papers, and provides detailed annotations for evidence localization and reasoning traces, together with a unified evaluation protocol. We assessed 15 models across 24 input configurations and conducted a fine-grained analysis of MLLM capabilities for all error categories. Across the board, retrieval-augmented generation (RAG) methods yield no significant improvements, revealing systematic deficiencies of current MLLMs on scan-oriented tasks and underscoring the challenge posed by ScholScan. We expect ScholScan to be the leading and representative work of the scan-oriented task paradigm.

Rongjin Li, Zichen Tang, Xianghe Wang +9

Beijing University of Posts and Telecommunications

LLM Evaluation Scientific Reasoning in Language Models Multimodal Reasoning

Ask PDF

Jul 30, 2026

AutoSupervision: Closing the Feedback Loop in Scientific Workflows with Grounded Revision Verification

Recent advances in large language models (LLMs) have enabled AI systems to assist scientific research and peer review. However, an essential capability for reliable AI-assisted scientific workflows remains underexplored: verifying whether reviewer feedback leads to meaningful and evidence-supported manuscript improvements. We introduce AutoSupervision, which evaluates whether scientific manuscript revisions genuinely address reviewer concerns through grounded evidence. AutoSupervision leverages transparent peer-review records as a natural source of supervision, where reviewer comments specify scientific concerns, author responses describe claimed resolutions, and revised manuscripts provide evidence of changes. Given reviewer comments, author responses, and revised manuscripts, models must characterize reviewer concerns, determine whether concerns have been addressed, and identify supporting manuscript evidence. We construct AutoSupervision from 56,000 Nature Communications articles and corresponding review records. Then we conducted experiments on LLMs, the ablation study, and the case study. Our results show that while LLMs perform well in characterizing reviewer concerns, with GPT-5.5 achieving a score of 0.754, evidence-based verification remains the primary bottleneck, with the best-performing model reaching only 0.501.

Haobo Li, Eunseo Jung, Wenxiao Zhao +8

1Shanghai AI Laboratory · 3The Hong Kong University of Science and Technology · University of California, Los Angeles

LLM Evaluation Scholarly Peer Review Automated Peer Review

Ask PDF

Jul 17, 2026

ToolSciVer: Multimodal Scientific Claim Verification with Visual Tool Augmented Reinforcement Learning

Multimodal Scientific Claim Verification (MSCV) requires models to verify scientific claims using visually grounded evidence from papers, including figures, tables, charts, and textual context. However, existing methods often fail because they struggle to locate decisive visual evidence, accurately read structured scientific visuals, and integrate multimodal observations into reliable reasoning. We introduce ToolSciVer, the first tool-augmented framework for MSCV to our knowledge. ToolSciVer equips a VLM with three type-aware visual tools, table row/column focus, chart-to-structure parsing, and high-resolution region zoom, which convert dense scientific visuals into explicit, claim-facing evidence, and trains the policy with Group Relative Policy Optimization (GRPO) under a composite reward of answer correctness, format validity, length control, tool-use efficiency, and tool-validity penalties. Experiments on SciVer and MuSciClaims datasets on five VLMs from three model families (Qwen, InternVL, Gemma) demonstrate that our method achieves superior performance compared to four competitive baselines including prompting-based and RL-based tool-use methods, highlighting the effectiveness of learned, type-aware tool use for scientific claim verification.

Binglin Zhou, Peng Shi, Ryo Kamoi +2

1The Pennsylvania State University · University of Waterloo

Tool Use in VLMs Scientific Claim Verification Multimodal Tool Use

Ask PDF

Abstract

Language Model Error Detection

Explore similar work

Not Search, But Scan: Benchmarking MLLMs on Scan-Oriented Academic Paper Reasoning

AutoSupervision: Closing the Feedback Loop in Scientific Workflows with Grounded Revision Verification

ToolSciVer: Multimodal Scientific Claim Verification with Visual Tool Augmented Reinforcement Learning